extract-function

name	extract-function
description	Model-only refactoring playbook for pulling a coherent block out of a long function into a named helper.
argument-hint
user-invocable	false
disable-model-invocation	false

extract-function

A model-only refactoring playbook. There is no slash command; the model invokes this to pull a coherent block out of a long function into a named helper.

When to reach for it

• A function spans more than one screen and reads as several phases.
• A comment introduces a block ("# now normalise the rows") — the comment is begging to become a function name.
• The same computation appears in two places (DRY).

Canonical procedure

1. Identify the block and the minimal set of variables it reads (inputs) and the single value it produces (output). If it produces two, that is two extractions.
2. Name the helper for WHAT it returns, not how — normalised_rows, not do_step_two.
3. Lift the block into a pure function where viable: explicit args in, explicit value out, no hidden state mutation.
4. Replace the original block with a call; keep the surrounding function's shape so the diff is reviewable.
5. Give the new helper full type hints, a docstring with a Raises: block if it can raise, and boundary + error-path tests if it is public.

Guardrails

• Stop and reconsider if the helper needs four or more parameters — that often signals a missing value object, not a missing function.
• Three similar lines are fine; do not extract a one-line helper used once (YAGNI / KISS).
• Use named arguments once arity reaches three (explicit-over-implicit).